Denoising Diffusion Probabilistic Model (DDPM)
Definition
DDPM refers to a generative model that can create images from a noise process.
DDPM is a class of generative probabilistic models for image synthesis. The learned reversal of a noising process enables generating a new image from noise.
Simply explained
Imagine an image full of noise gradually forming a motif step by step. This is an analogy for generation, not the restoration of an actually existing photo.
Why it matters in practice
DDPM is a foundational approach in generative image AI. Model quality and execution costs must be evaluated based on a specific use case.
Example from business
Possible example: A research team compares image samples from different diffusion models using uniform quality metrics.
Limits and pitfalls
- Generated images do not depict real events. Quality and runtime depend on training data, model, and sampling method.
Origin & history
Jonathan Ho, Ajay Jain, and Pieter Abbeel published the first arXiv version of their DDPM paper on June 19, 2020.
Use in mid-sized companies
- Possible applications include feasibility prototypes for image synthesis and comparison of different generative model architectures.
Frequently asked questions
Is DDPM an image editing filter?
No. It is a generative model.
Is DDPM identical to DDIM?
No. DDIM addresses a different class of sampling processes.
Sources
- Denoising Diffusion Probabilistic Models · Ho et al. (2020): Denoising Diffusion Probabilistic Models
- Denoising Diffusion Implicit Models · Song et al. (2020): Denoising Diffusion Implicit Models
Sources checked on 20.10.2026. Editorial use cases are hypothetical; no evidence of real customer results.
Editorially reviewed on: